Leuven: The AI Act high-risk deadlines have moved, and compliance teams across Europe are rewriting their plans. Under the Digital Omnibus on AI, agreed politically on 6 May 2026, stand-alone high-risk systems now face 2 December 2027, while AI built into regulated products gets until 2 August 2028.
What the new AI Act high-risk deadlines say
The original text set 2 August 2026 for Annex III systems such as recruitment tools, credit scoring and education software. Annex I systems, which cover AI inside medical devices, machinery and vehicles, faced 2 August 2027. The Commission first proposed the Digital Omnibus on 19 November 2025 with a conditional trigger tied to the availability of standards.
Negotiators dropped that trigger and wrote fixed dates into the law instead. Fixed dates give companies certainty, and they remove the argument over whether standards are ready. A first trilogue round on 28 April ended without a deal, and the political agreement followed eight days later.
Which obligations still apply sooner
Delay does not mean a pause on everything, and the AI Act high-risk deadlines cover only one part of the law. The Article 50 transparency duties still apply from 2 August 2026, although systems already on the market get until 2 December 2026 to meet the watermarking rule in Article 50(2). The agreement also adds a new Article 5 ban on AI systems that generate non-consensual intimate imagery and child sexual abuse material, with a transition period to 2 December 2026.
Providers should therefore treat the calendar as split. Prohibitions and transparency duties arrive first, and the heavy conformity work for high-risk systems follows later.
How governance shifts to the AI Office
The agreement strengthens the EU AI Office. It gains exclusive competence over certain AI systems built on general-purpose models from the same provider, and over AI systems integrated into very large online platforms. It also receives investigation, inspection and fining powers.
National authorities keep competence for law enforcement, border management, judicial authorities and financial institutions. This division concentrates oversight of the largest model providers in Brussels while leaving sector supervision with national regulators.
Winners, losers and the cost of waiting
Delay changes incentives. Firms that planned to finish conformity work by August 2026 now face a longer runway, and the extra time lowers the cost of getting the paperwork right. The AI Act high-risk deadlines also give standards bodies space to finish technical specifications that were running late.
Industry groups asked for more time, and they received it. Producers of machinery are among the biggest beneficiaries, because the text largely exempts AI in products under the Machinery Regulation and narrows the definition of a safety component. Providers of recruitment and credit tools gain sixteen months.
The cost falls on workers and consumers, who wait longer for the safeguards that the AI Act high-risk deadlines were meant to deliver. Companies that pause preparation also risk a compression of the work into 2027, when standards, notified bodies and national authorities all need to be ready at once.
What to do before 2027
Organisations that build or buy AI should keep their inventories of AI systems current and map each one to Annex III or Annex I. They should also use the extra time to test bias detection, since the omnibus widens the legal basis for processing special-category data for that purpose under a strict necessity standard. Treat the new AI Act high-risk deadlines as a window for preparation, not a reason to stop.
